Artificial intelligence is still far from replacing the programmers directly

1. Overview

Gen Ai and tools like Github Copilot are not good in replacing programmers for dangerous tasks, starting from April 2025. We seem to be still several months away from that stage. a lot Data seen on the Internet Either Marketing campaigns Try to sell the current artificial intelligence tools, or Very optimistic data is not based on experience With the actual tools present today.

In this article, we offer a summary of three online articles to support the above statement, and the latter is the experience of the personal development of this author with GitHub Copilot.

2. Article 1: Artificial intelligence agents cannot reliably correct correction programs.

Article 1: The researchers say artificial intelligence is not ready to replace human programmers to correct errors

summary:

• Even when accessing tools is available, the artificial intelligence agent cannot reliably correct correction programs.
• .. (People) are their expectations because the models are not good enough in the error correction part, and the correction of errors occupies most of the developer’s time. This is it A proposal from Microsoft Research.
• Artificial intelligence models … crying out of what an experienced human developer can do
• … suggested some ambitious ideas about artificial intelligence agents who directly solve the developers far from reality
• … models tend to produce a symbol loaded with imbalance and security gaps, and they are generally unable to fix these problems
• The best result is the factor that keeps the human developer a great deal of time, not the result that can do everything they can do

3. Article 2: Under pressure to embrace artificial intelligence, developers grow in frustration

Article 2: Artificial intelligence coding details pushing developers to the edge of the abyss

summary:

• ..Ii adopts “tearing their company” with the emergence of a dispute between the leadership and the employees who adopt such tools.
• For software developers specifically, there are fears that artificial intelligence coding tools make errors in their software instructions, have failed in many tasks, and more technical debts.
• … the developer’s belief in artificial intelligence coding tools has decreased quickly
• … developers describe a large number of technical problems and headaches associated with artificial intelligence coding tools, from how they repeatedly suggest an incorrect symbol and even delete the current code to many problems that cause them with publishing operations
• … The use of artificial intelligence tools causes an increase in accidents, as 68 % of respondents say that they spend more time solving the AI ​​security weaknesses now before using artificial intelligence coding tools
• GitHub Copilot … It was wrong several times as it was true
• It takes a lot of time for someone (human) to review something (artificial intelligence tools) they did not write and understand completely
• Executive managers weaken their expectations … the company’s leaders who do not have a close vision of the engineering workflow
• Artificial intelligence coding tools are still new, and know how to use it effectively requires a coordinated effort

4. Article 3: Github Copilot (GHC) cannot be trusted with a bit complex task

Article 3: GitHub Copilot (Gen-Ei) is useful, but not great

summary:

• GitHub Copilot (GHC) … Sometimes it is great .. Sometimes your time is lost, especially because the long answers you provide are often outside the subject
• It is sometimes useful, but only for local domain problems, and he cannot see the largest image.
• … “Personal feeling” is “does not know it well”, he “is trying to guess”, and because it is a huge memory machine of millions of code preserved code, the guesses are sometimes wonderful, and sometimes outside the subject.
• .. a great disappointment that the sentence C# construction can not be obtained correctly all the time and check the presence/methods of C#
• … the active line of the symbol was deleted because a similar line of software instructions was suspended
• I use a “ghost test” from GHC a lot, review it, and accept suggestions when I love them.
• I use the GHC Prompt to request the generation of cuts or small jobs with clear functions …. Any guarantees for GHC will not succeed here, but if that happens, it may be great.
• I no longer try more changes that include 3-4 files at the same time … Answers at best are incomplete, with many errors, such as C# and methods that do not exist (hallucinations?)
• … it’s time and consumes energy review every answer provided by GHC
• … cheerful for C# features and methods that are not found in well -known and used applications for applications …
• … generates more chaos than useful software instructions, for the specific tasks based on patterns that include 4 files.
• Ask to help from GHC in problems, but read MAX 2 chats. Its answers tend to be long … If you don’t give me a good answer in two attempts, I will go to Google for the same problem.
• GHC has a serious concentration problem. Answers often outside the subject.
• Github Copilot (GHC) cannot be trusted with a somewhat complex task that includes several files at the same time. In such scenarios, the results are incomplete and not effective for time compared to direct manual programming.
• GitHub Copilot (GHC) tends to hallucinations around the methods and characteristics of the non -exist. GHC-
• The created code is not assembled immediately, which requires a lot of manual work to end it.

5. Conclusion

The above articles indicate that some ambitious ideas about artificial intelligence tools that replace developers are directly out of reality.

Some authors believe that it is likely to be the best result is the Amnesty International tool that saves the human developer a great deal of time, not those that can replace the developer directly.

6 references

[1] The researchers say artificial intelligence is not ready to replace human programmers to correct errors

https://arstechnica.com/ai/2025/04/researsers-find-ai-is-pretty-bad-at-atbugging-but-heyre-working-on-et/

[2] Artificial intelligence coding details pushing developers to the edge of the abyss

https://leaddev.com/culture/ai-coding-Mandates-are-driving-developers-to-the-rink

[3] GitHub Copilot (Gen-Ei) is useful, but not great (March 2025)

https://markpelf.com/2717/github-copilot-gen-ai-is- heelpful-but-not-great-march-2025/